Structured Bayesian learning for recovery of clustered sparse signal

Lu Wang, Lifan Zhao, Lei Yu, Jingjing Wang, Guoan Bi

Abstract: Highlights•This paper considers the problem of recovering clustered sparse signals with unknown cluster sizes and locations by proposing a new sparse Bayesian learning algorithm. This algorithm can be regarded as a generalization of both pattern-coupled sparse Bayesian learning (PC-SBL) and Clustered Sparse Solver (Cluss).•The proposed algorithm separates the sparse coefficient into support and amplitude on which structured priors are imposed. A generalized pattern-coupled prior is adopted for the coefficient amplitudes by introducing a local structure indicator, which allows coupling function to vary automatically against the local inherent data structure in a probabilistic manner. The prior of the support is also probabilistically selected by three different patterns defined on the local structure of the data. The two structured priors perform their functions coherently with each other on the sparse coefficient.•Furthermore, an adaptive support utilization strategy is adopted to continuously update the support according to its credibility. This strategy greatly reduces the utilization of the inaccurate information of the support when there are less measurements available.

External IDs:dblp:journals/sigpro/WangZYWB20